Preprint
Streaming End-to-end Speech Recognition For Mobile Devices
Yanzhang He et al., arXiv (ICASSP 2019), 2019
Read the originalRead on October 17, 2026
- What it is
- Google researchers describe building an end-to-end speech recognizer with a recurrent neural network transducer that is small enough to run on a phone. They compare it with a conventional CTC-based model on latency and accuracy.
- How much weight it carries
- Preprint from the team that shipped an on-device recognizer, and listed as an ICASSP 2019 paper. The comparison is with Google's own models on Google's own test sets, so the size of the gain is not independent evidence for other systems.
Facts taken from this source
- He and colleagues report that their RNN transducer can outperform a conventional CTC-based model in both latency and accuracy across a number of evaluation categories.
“we find that the proposed approach can outperform a conventional CTC-based model in terms of both latency and accuracy in a number of evaluation categories” (Abstract)